Wireless Networks for Smart Surveillance:Technologies, Protocol Design and Experiments

M. Spadacini1,2, S. Savazzi1, M. Nicoli1, S. Nicoli21DEI, Politecnico di Milano, Italy. 2Comelit S.p.A. Bergamo, Italy.

E-mail: (savazzi,nicoli)@elet.polimi.it, (milo.spadacini,sergio.nicoli)@comelit.it

Abstract—Wireless machine-to-machine (M2M) networks arebecoming a relevant topic in the field of home automationand advanced security systems. A wireless network for indoorintrusion detection is based on several sensors that are deployedover the monitored area for detecting possible risky situations andtriggering appropriate actions in response. The network needs tosupport traffic patterns with different characteristics and qualityconstraints. Namely, it should provide a periodic low-powermonitoring service and, in case of intrusion detection, a real-timealarm propagation mechanism over inherently unreliable wirelesslinks subject to fluctuations of the signal power. Followingthe guidelines introduced by recent standardization, this paperproposes the design of a wireless network prototype able tosatisfy the specifics of the intrusion detection application. Aproprietary medium access control is developed based on the low-power SimpliciTI radio stack. Network performance is assessedby experimental measurements using a test-bed operating at868MHz in an indoor office environment with severe multipathand non line-of-sight propagation conditions.

I. INTRODUCTIONWireless machine-to-machine (M2M) technologies are gain-

ing the attention of companies operating in the field of homeautomation and advanced security systems. Although mostsolutions currently on the market are still based on power-lineor wired communications, it is expected that cable replacingwill eliminate time-consuming installations [1], opening theway to more flexible services for the end-user. The gen-eral requirements of wireless intrusion detection systems areevaluated at European level through the standard EN 5013-5-3 [2] that regulates several aspects such as immunity tothe variations in the channel, collisions, intentional (or not)replacement of devices and interference. It also provides speci-fications for link monitoring, antennas, and test procedures forthe verification of these properties. For each of these itemsthe legislation defines four classes corresponding to differentlevels of security.A wireless network for home automation is composed

of a number of wireless-embedded sensors and actuators(e.g., infrared, motion sensors, light switches, safety sen-sors, accelerometers) that intelligently share their transmis-sion resources and interconnect with each other accordingto a suitable wireless architecture. As depicted in Fig. 1, anetwork coordinator or Access Point (AP) is responsible forstarting the network, choosing and monitoring the frequencyfor transmission and reception, joining new End Devices (ED).ED nodes can be spread over the area of interest (typicallyindoor and over different floors) while communication with

AP Controller

ED: End device

Fig. 1. Floor plan map and WHAN deployment with an AP controller and8 EDs.

the AP coordinator might be guaranteed by Range Extender(RE) devices serving as decode and forward relays.In this paper, we consider the design and implementation

of an indoor wireless network for indoor surveillance andintrusion detection. The system needs to periodically recordactivities in the environment and provide a highly reliableconnection service for alarm message propagation. Severalexamples of Wireless Home Automation Networks (WHAN)have been discussed in the literature. A Bluetooth WHANhas been introduced in [3] using a primary network controllerand a number of sub-controllers connected by star topology.However, the wireless architecture does not completely replacecabling and the use of the Bluetooth technology shows disad-vantages in terms of access delay. A ZigBee-based WHANhas been proposed in [4]. Although ZigBee interface basedon the IEEE 802.15.4-2006 standard [5] provides an effectivenetwork solution for low-power wireless sensing, the overallsize of the radio stack (between 45 to 100 kbyte) limitsits applications to a small subset of smart home automationscenarios.Contribution of the paper. This paper focuses on new

promising wireless technologies for next-generation smartsurveillance systems and presents a network design validatedby experimental tests. A proprietary Medium Access Control(MAC) link-layer protocol tailored for intrusion detection hasbeen developed on top of the SimpliciTI compliant radiostack [6] based on the ultra-low-power CC430 microcontrollerSystem-on-Chip (SoC) with integrated sub-GHz RF trans-

WCNC 2012 Workshop on Internet of Things Enabling Technologies, Embracing Machine-to-Machine Communications andBeyond

978-1-4673-0680-5/12/$31.00 ©2012 IEEE 214

Keep Alive message framing structure

Fig. 2. Frame structure and keep-alive message passing.

ceiver core and low-power MSP430 controller [7]. SimpliciTIwireless modules use a very basic core API compared toZigBee and thus allow for a more flexible network design. Theproposed MAC sub-layer has a smaller code size (8-16kbyte)compared to the ZigBee radio stack and it allows to efficientlymanage the different traffic patterns generated by the intrusiondetection system, jointly exploiting both synchronized andnon-synchronized access schemes. The test-bed developed forthis study uses radio devices operating at 868MHz ISM band;configuration and planning of the WHAN have been carriedout in an indoor environment with severe multipath and nonline-of-sight (NLOS) propagation.

II. NETWORK REQUIREMENTS FOR SMARTSURVEILLANCE SYSTEMS

A WHAN designed for smart surveillance and intrusiondetection has many peculiarities that are common to a broaderclass of automatic industrial control systems [8]. The physical(PHY) layer radio characteristics that are mostly common toall these systems are low data-rate (below 500kbps), carrierfrequency at 2.4GHz or in the 915MHz/868MHz ISM bands,and receiver sensitivity above −100dBm [9]. Most PHY layeraccess schemes are based on OQPSK or FSK combined withDirect-Sequence Spread Spectrum (DSSS) transmission, asproven technologies that reduce the network susceptibilityto interference, providing a few dBs of coding gain andsome improvements over harsh environments characterizedby multi-path fading. Multi-channel radios are also adoptedto efficiently manage co-channel interference and reject anyexternal disturbance through dynamic scheduling and chan-nel/frequency hopping.Basic requirements that need to be considered for the

WHAN protocol design are listed in the following:Network services. A wireless network for smart surveillance

must support different classes of traffic patterns with relatedquality constraints. The network should provide both a peri-odic low-power monitoring service and a real-time alarm prop-agation mechanism which must be robust enough to cope withinherently unreliable wireless links characterized by signalpower fluctuations. Intrusion detection systems have stricterreliability and delay requirements compared to conventionalhome automation services. Reliable communication occursonly if both sensor observations and feedback from the AP

controller are decoded by respective parties within specifieddeadlines defined by the controller policy. This hard constraintcalls for an advanced wireless link-layer protocol managementto provide an optimal trade-off between reliability and real-time communication.Indoor radio planning. Indoor home environments are typ-

ically characterized by severe multipath due to the presenceof reflective surfaces (e.g., walls, floor, furniture) [9]. Radioplanning is a useful tool which relies on the prediction of wire-less link quality. Prediction can be supported by independentradio measurement campaigns over typical indoor buildingand/or by empirical propagation models. A low accuracy inthe radio planning design phase will turn into high logisticcosts: adding new wireless range extenders to improve thecoverage as well as moving them around the environmentmay become unacceptable and highly time-consuming in somecases. For this reason, it is crucial to develop design guidelinesand tools that can allow to achieve a reasonable accuracy inthe prediction of the wireless coverage.Low duty-cycling operation. Wireless autonomous devices

are battery-powered: EDs are usually deployed in predefinedspots and must remain active for 3-5 years. This poses strin-gent constraints on the sensor and radio transceiver design forminimum energy consumption. The MAC layer protocol needsto be optimized to preserve the battery. Energy harvestingtechniques also provide a powerful tool for lifetime maxi-mization. Some of the techniques employed to reduce powerconsumption includes: i) dynamic sleep mode activation (withfast wake-up times) to shut-down devices when not transmit-ting or receiving; ii) low duty-cycling design to minimize EDactivity cycles. Being the network almost static, the adoptionof guaranteed (interference-free) time-division multiple access(TDMA) and beacon-enabled network designs [10] are to bepreferred compared to random access strategies to minimizeidle listening.

III. WHAN ARCHITECTURE AND MAC DESIGNThe proposed network architecture, depicted in Fig. 1,

consists of an AP and a number of REs and EDs.The AP is the network coordinator which manages a low-

power radio interface for two-way communications with theremote EDs and acts as a translator over any external network.From the application layer, the AP node should guarantee the

215

Fig. 3. Message exhange during channel hopping phase.

interoperability of the wireless infrastructure with other end-user operator services, i.e., portable human machine interfaces(HMI), radio-frequency identification (RFID), video-camerasor thermocameras monitoring. As a consequence, the APshould perform the following tasks: i) adaptively choose thenetwork resources (through channel hopping and dynamicpower control); ii) register new EDs (joining phase) andsynchronize them to guarantee low duty-cycle activity as pre-scribed by the standard [2]; iii) periodically monitor the devicestatus through standard-compliant keep-alive messages; iv)guarantee real-time alarm message propagation with minimumlatency.The RE is the device responsible for multi-hop communi-

cations. Its basic function is to repeat the message from theED under its control to the AP and viceversa.Finally, the ED is the low power input/output wireless

instrument that interacts with the sensor hardware to monitorthe indoor environment and detect intrusions. Any ED shouldperform two functions: i) periodic transmission of keep-alivemessages containing basic information on radio device status(battery residual levels, receiver sensitivity, channel quality),sensor status and tampering; ii) transmission of alarm mes-sages within a maximum latency of ∼ 10s. Periodic re-transmission of keep-alive messages conforms to the standardGrade 2 [2] that prescribes a minimum refresh rate of 20min.

A. Network ConfigurationThe network configuration phase allows to register new-

comer devices. Devices have to inform the AP about the role,

Time between wake up and beacon reception [ms]

Wake up

Wake up

tΔ

tΔ

Probability density function (pdf)

Fig. 4. Top figure: device synchronization algorithm assuming two cases:wake up before beacon reception (case 1), wake up after beacon reception(case 2). Case 2 requires re-synchronization using the following beacontransmission. Bottom figure: probabilty density function of the time betweenthe ED wake up and the reception of beacon message (avg 50ms equals toselected guard-time Tg , maximum jitter is 4ms).

the type and the number of their sensors in the network.A device can join the network either with direct or indirect(through a RE) connection. A RE can join the network onlywith a direct connection while an ED can exploit either director indirect connection. In case of indirect connection the EDcommunicates with a RE that forwards the message to/fromthe AP. Routing of packets is established at the time of networkconfiguration.

B. Frame Structure and Traffic ManagementThe proposed MAC sub-layer defines a proprietary frame

structure that jointly handles the periodic traffic needed formonitoring the status of the sensing devices and the potentialbursty traffic generated by alarm messages in case of intrusiondetection. The first task is implemented through a keep-alivemessage exchange with dedicated channel assignments; thesecond one by random slotted access through Carrier SenseMultiple Access with Collision Avoidance (CSMA/CA).The frame structure is shown in Fig. 2. Time division duplex

is employed to separate uplink and downlink. Logical frameshave length Tframe = 208s and consist of 64 time-slots ofTslot = 3.25s separated by guard times. The frame durationdepends on the number of EDs in the network, here 64.Periodic keep-alive message exchange session (of duration

Tframe) starts with the transmission of the first keep-alivemessage from ED #1 (time slot 1) and stops with the lastfrom ED #64 (time slot #64). Each time-slot starts withthe AP beacon message that contains the identifier (TurnInformation or TI) of the device that has to update itsstatus information. The beacon message is used for devicesynchronization following a similar approach as in [11] (seeSec. III-D). The time-slot is further subdivided into 65 mini-slots of duration T = 50ms, the first one being reserved forkeep-alive message transmission (TI Answer) of the deviceindicated in the beacon message (holding the turn). A two-

216

way explicit acknowledgement policy (see Fig. 2, left-side andTI-ACK message) is adopted to guarantee reliable messagedelivery over fading links. The remaining 64 mini-slots areused for synchronization operations and for alarm propagationas indicated below.For real-time alarm delivery, any ED detecting an intrusion

cannot wait its reserved mini-slot, it has to propagate the alarmmessage immediately after the detection. The solution hereproposed is to use a random access over the above-mentioned64 mini-slots. The ED intending to give the alarm overhearsthe beacon message (even if intended for another ED) toacquire synchronization, randomly chooses one of the 64 mini-slots and uses carrier sensing before any transmission attemptto avoid cross-tier interference from external devices operatingon the same frequencies (e.g., WiFi, Bluetooth). In the worst-case the overall latency of alarm message propagation is thuslimited to two times the slot duration 2Tslot = 6.5s. If thefirst alarm transmission fails due to collision, the ED waits arandom period (in number of mini-slots) before retransmission.If also the second-one fails the ED checks if the AP had changechannel by performing two retransmissions for each availablechannel until an ACK is observed.

C. Interference-aware Channel HoppingTo avoid bursty errors over consecutive polling sessions,

a channel frequency hopping (CH) phase can be adaptivelyinitiated by the AP controller. CH is commonly adopted inindustrial communication as it provides increased robustnessagainst interference and some additional protection againsteavesdroppers. In the proposed system the chosen carrierfrequencies are 868 MHz and 869 MHz, respectively. Asshown in the message exchange example of Fig. 3, the APcontinuously monitors the background noise of the radiochannel and it might decide to perform a channel hopping incase of severe interference. At the first beacon signal lost, theED programs a timer at 2Tslot. If after the timer interval theED does not receive the next beacon signal, the ED changesits working frequency accordingly. This waiting period isnecessary to be sure that beacon miss-detection is not causedby a temporary interferer.

D. Duty Cycling and Device SynchronizationDevice synchronization is based on the approach in [10]. A

drift error compensation algorithm is developed to minimizeidle listening and thus maximize the device lifetime. EachED sleeps during the beacon transmissions intended for otherdevices and, excluding the alarm transmissions, it wakes-uponly to reply to its intended beacon message. The choseninterval Tframe among successive beacon messages is largecompared to typical re-synchronization intervals of 10 − 15s[10]. Time-synchronized duty cycling is thus guaranteed bycorrecting and updating the sleep time Tsleep (the time elapsedbetween two consecutive wake-ups) on every new beaconmessage reception to account for the timing error experiencedwith the AP local oscillator. The proposed synchronizationalgorithm allows the ED to turn on the radio Tg = 50ms beforethe beacon message reception, being Tg a pre-defined guard

time. The residual timing uncertainty after drift correctionmakes the observed interval ∆t between the ED wake-up andthe reception of the intended beacon message to be modeledas Gaussian distributed random variable with mean Tg =50ms and with maximum jitter of 4ms. Observed probabilityfunction (pdf) is shown in Fig. 4 at bottom. Given that thek-th beacon message is received, the ED device updates thesleep time Tsleep(k) before the next wake-up by following aLeast Mean Square (LMS) approach as

Tsleep(k) = Tsleep(k − 1) + μ(∆tk − Tg) (1)

where ∆tk = ∆t(k) is the observed time between theED wake-up and the reception of k-th beacon message.Initial value for Tsleep assumes perfect synchronization asTsleep(0) = Tframe − T − Tg, updating factor μ is chosenhere as μ = 1/2. As demonstrated in the experimental activity(see Sec. IV), a larger guard time (Tg > 50ms) would causeunnecessary idle listening and higher energy consumption.The LMS tracking algorithm (1) can be applied when the

intended k-th beacon message is correctly received: in Fig 4-top this scenario is referred to as Case 1. Case 2 refers insteadto a scenario where the ED loses the beacon transmission dueto a radio interference or a residual clock drift that delays itswake-up. In this case the LMS tracking is not applied whilean ad-hoc re-synchronization policy is implemented. The EDstays awake to receive the next beacon transmission (intendedfor ED holding the next turn). Next, to avoid interferencewith the transmission of the ED indicated in the beacon, itsends the keep-alive message using a reserved mini-slot chosenamong the available 64 mini-slots that follows the first mini-slot reserved for keep-alive message exchange1 (see framingstructure in Fig. 2.

E. Dynamic Transmission Power AllocationA dynamic transmission power control algorithm is imple-

mented to minimize the energy expenditure during periodickeep-alive message transmission. For each ED the AP con-troller measures the received signal strength indicator (RSSI)from the received TI Answer message. The measured RSSIvalue Γ is embedded into the TI ACKnowledge message andthus sent back to the corresponding ED. The ED uses thisinformation to adapt the transmit power level PT for thenext keep alive message transmission. The goal of the powercontrol algorithm is to adaptively adjust the power level PT sothat the RSSI measurement Γ can be kept below a maximumthreshold Γmax and above a minimum threshold Γmin. Thethreshold used during the experimental activity (Sect. IV) areΓmax = −20dBm and Γmin = −70dBm.

F. EncryptionTo improve the security level all the transmissions are coded

with the XTEA encoding scheme [12]. It is composed by threemain elements which are a 128-bit encryption key, a 32-bitinitialization vector and a 32-bit counter. The initialization1Reserved mini-slot for re-synchronization is assigned during the network

configuration phase.

217

Energy consumption profile

Fig. 5. Left-side sub-figure: measured absorbed current (mA) during wake-up, keep-alive RX and TX phase (transmission of ACK message is included). Right-side sub-figure: battery lifetime versus the number of transmitted alarm messages per hour. Guard-time of 50ms uses the proposed drift error compensationfor prediction of timing uncertainty, the case of guard time 150ms does not implement any error compensation.

vector and the encryption key are set at built-time while thecounter value is determined at the time of the link creationbetween two or more devices. Devices that have formed thelink preserve independent counters.

IV. EXPERIMENTAL ACTIVITY AND RESULTSWireless modules used for test-bed and experimental ac-

tivity are part of the EM430F6137RF900 Texas Instrumentkit with integrated System-On-Chip RF XCC430F6137IRGC.Key feature of the SOC are the follows: i) energy consumptionfrom data-sheet is characterized by a RX absorbed current of15mA, while TX current is 33 mA for PT = 12dBm (max.RF transmit power); ii) radio transceiver module is CC1101with minimum sensitivity -111dBm, data-rate 500 kbit/s and2-FSK as RF modulation; iii) microcontroller (MCU) core isMSP430 with AES compliant coprocessor, flash memory is 32kB, RAM 4 kB, ADC 12bit. Experimental activity has beenconducted in an indoor environment consisting of 8 adjacentrooms. A single wireless ED was deployed in each roomto monitor the surrounding area. People were moving insideeach room causing random fluctuations of radio signals. Forall devices the antenna height from ground was below 1m.Propagation took place over an harsh radio environment withmetallic objects (e.g., coaxial cabling, monitors/PCs, tubesfor air conditioning, etc.) and furniture causing additionalattenuation.

A. Analysis of network lifetimeFigure 5 shows the energy consumption profile of a battery-

equipped ED during a normal cycle of remote control. Mea-surements were obtained through an oscilloscope connectedin parallel with the node itself. The absorbed current hasbeen observed during five different phases: the wake-up period(approx. 10mA), the beacon message reception (23mA), thestate message transmission (40mA), the acknowledge recep-tion (24mA) and the sleep mode (8μA). To avoid the highbackground noise of the oscilloscope, the absorbed currentduring the sleep period was measured by a precision ammeter.

Values measured during reception and sleep mode differ fromthe declared power consumptions, respectively 15mA and5μA. In sleep mode, the difference is due to the consumptionof the wake-up timer not considered in the data-sheet.Given the requirements of Grade 2 service, the remote

control by keep alive message exchange is repeated about 17times per hour (3600/Tframe), therefore the ED battery lifetime assuming a guard time of Tg = 50ms can be estimated to9 years using a commercial 1.4Ah lithium battery. In Fig. 5,the impact of the alarm transmission rate (expressed in averagenumber of alarm messages per hour) and of the choice for theguard time on the expected lifetime of ED nodes are analyzed.We considered two cases: in the first one drift compensationis employed according to the specifics of the proposed MACsub-layer with guaranteed guard-time of Tg = 50ms; thesecond case refers to the use of a guard-time of Tg = 150mswithout clock drift compensation. The use of the proposeddrift compensation strategy allows significant gains in networklifetime (up to two times for alarm transmission rate below 200messages/hour).

B. Channel measurements and network planningIn this section, we analyze the impact of radio propagation

at 868 MHz on the wireless alarm message delivery. For theexperimental activity we used 8 EDs deployed in differentrooms and one AP controller. The effects of propagationin each office room has been characterized by calculatingthe average and the standard deviation of the RSSI samplesmeasured over each AP-ED link. The radio environment hasbeen observed over a period of 48 hours during which theposition of the AP was fixed while the EDs were movedin two different locations in each room on every 24 hoursof operation. To highlight the impact of channel fluctuationstransmit power is here fixed to PT = 12dBm while dynamicpower allocation is disabled.The resulting values of RSSI average and standard deviation

are shown in Fig. 6 and Fig. 7, respectively. From the compar-ison between daytime (top sub-figure) and night (bottom sub-

218

Daytime

Night

Fig. 6. Average RSSI measured in different rooms during day (top figure)and night (bottom figure). RSSI samples are averaged over a measurementperiod of 12 hours.

Daytime

Night

Fig. 7. Standard deviation of the RSSI measured in different roomsduring day (top figure) and night (bottom figure). Red rooms indicate criticalenvironments (largest RSSI deviation) for network deployment.

figure) it is easy to notice that although the observed averageRSSI does not change significantly from daytime to night, thestandard deviation is strongly influenced by the presence ofpeople in the propagation environment (during daytime [13]).The behavior of channel fluctuations is thus non-stationary asthe standard deviation in some rooms is shown to increasefrom about 0dB during the night to more than 4dB duringdaytime. As also observed in [14], this suggests to dynamicallyincrease transmit power during daytime by adding a 4dBconstant fade margin to compensate for the random signalstrength fluctuations. During night, transmit power can beinstead decreased by 4dB on average.

V. CONCLUDING REMARKSIn this paper a proprietary MAC link-layer protocol tailored

for smart surveillance and intrusion detection applications hasbeen developed on top of the SimpliciTI compliant radio stack[6] using the ultra-low-power CC430 microcontroller [7]. Theproposed protocol uses a very basic core API compared toZigBee, allowing for a more flexible network design. Themedium access scheme has been designed to jointly exploitboth scheduled and random access to guarantee low-powerperiodic keep-alive message exchange and real-time alarmpropagation with minimum latency, respectively. The analysisof battery consumption proved that low-power duty cyclingcan guarantee a lifetime of several years. Network planning inan indoor environment and power control are also discussed.

REFERENCES[1] A. J. D. Rathnayaka, V. M. Potdar, S. J. Kuruppu, “Evaluation of

Wireless Home Automation Technologies,” Proc. of the 5th IEEE Int.Conf. on Digital Ecosyst. and Tech., Daejeon, Korea, June 2011.

[2] European Standard, EN 50131-5-3, “Alarm systems - Intrusion systems,Part 5-3: Requirements for interconnections equipment using radiofrequency techniques”, May 2005.

[3] N. Sriskanthan, F. Tan and A. Karande, “Bluethoot based home au-tomation system,” Microprocessors and Microsystems, vol. 26, no. 6,pp. 281-289, 2002.

[4] K. Gill, S.-H. Yang, F. Yao, and X. Lu, “A ZigBee-Based HomeAutomation System,” IEEE Trans. Consumer Electronics, vol. 55, pp.422-430, May 2009.

[5] Standard IEEE 802.15.4-2006, “Part 15.4: Wireless Medium AccessControl (MAC) and Physical layer (PHY) specifications for low-rateWireless Personal Area Networks (LR-WPAN)”, 2006.

[6] Larry Friedman, “SimpliciTI: Simple modular RF network specifica-tion,” Texas Instruments, Inc., San Diego, California USA, 2009.

[7] Data-sheet, “MSP430TM SoC with RF Core,” Texas Instruments, No-vember 2010.

[8] A. Willing, “Recent and emerging topics in wireless industrial commu-nications: a selection,” IEEE Trans. Ind. Inform., vol. 4, no. 2, May2008.

[9] C. Gomez, and J. Paradells, “Wireless home automation networks: asurvey of architectures and technologies,” IEEE Commun. Magazine,pp. 92-101, June 2010.

[10] W. Ye, et al. “An energy efficient MAC protocol for Wireless SensorNetworks,” Proc. IEEE INFOCOM 2002.

[11] S. Ganeriwal et al., “Estimating clock uncertainty for efficient duty-cycling in sensor networks,” IEEE/ACM Trans. on Networking, vol. 17,no. 3, June 2009.

[12] M. N. Roger and D. J. Wheeler. “Tea extensions,” Technical report,Computer Laboratory, University of Cambridge, 1997.

[13] P. Castiglione, S. Savazzi, M. Nicoli, T. Zemen, “Impact of Fading Sta-tistics on Partner Selection in Indoor-to-Outdoor Cooperative Networks,”Proc. IEEE Inter. Conf. on Comm., South Africa, May 2010.

[14] C. A. Boano, N. Tsiftes, T. Voigt, J. Brown, Utz Roedig, “The Impactof Temperature on Outdoor Industrial Sensornet Applications,” IEEETrans. on Industrial Informatics, vol. 6, no. 3, August 2010.

219